Multi-threaded N-body is back

Hello everyone,

We decided to once again provide support for the milkyway_nbody multi-threaded application. If you discover any issues with the new application, please do not hesitate to contact us so that we may expeditiously resolve them. Thank you all for your continued support.

-Eric

I request that you do one or the other, but not both, for the reasons previously discussed concerning the BOINC scheduler.
(I don't see any way to choose anything yet.)

I have tried 2 of the new jobs, both went off the rails
Am trying third one then will give up and do other projects for a while

---

Could you please clarify what you meant by "went off the rails?" Are they spitting out errors? Is the run-time too long?

Run time for me appears to be never-ending. After a while the percentage done stops advancing and the estimated time remaining starts climbing. Runs that were supposed to finish in 8 hours on 8 CPUs are still running after 2 days with over 1 day estimated time to completion. I've aborted anything estimated at over 5 hours to see if the shorter ones will complete.

One work unit estimated at under 4 hours is still running after 7:17 hours and stuck at 19.729% completed, estimated 1d 05:31 to completion (and climbing).

As far as I can tell my system has not completed a single non-GPU work unit since I restarted processing a week ago.

I have every CPU task finish without problems. CPU tasks seem to switch between multiple single processor and one multi processors without any problems at all on all five workstations as they progress down the task list.

for me the multi-threaded jobs says it will take 1-3 hours with 7 CPUs...however they have a real hard time finishing.. current job has been running 3:44...with an estimated 3:26 to go...however the estimated time is counting up instead of down. I dont have anything else running on my computer except for this browser....I have had to abort a ton of jobs as they pass their deadline. lots of wasted CPU it seems...

From what I'm seeing in my BOINC manager, the multi-threaded nbody application takes about 4 hours to complete on 8 CPUs. The main reason why these simulations take so long is due to two factors:
- We are using 40,000 bodies instead of 20,000. This is the minimum number of bodies we require to ensure the random seed does not drastically affect the final state of the nbody simulation. Our N-body algorithm is O(n log n), which means this change makes runs take 2.14 times longer than before.
- MilkyWay@home is optimizing to ultradense cores about 50% of the time. In order to accurately run a dense collection of bodies in an N-body simulation, you need to have a smaller timestep, otherwise, the collection of bodies explodes outwards. The denser the galaxy, the smaller the timestep needs to be. These dense progenitors take about 4 times longer to run than normal, and when MilkyWay@home optimizes to a dense progenitor, we end up with a population of parameters that each take several hours to compute.

While we cannot improve the number of bodies, we are working on removing runs that converge to these heavily cored progenitors. We apologize for the inconvenience. Thank you for your patience.

-Eric

I am running W10,. If the multi thread tasks are the ones using 6 cpus I am having trouble. The tasks run very slowly, 3% after two days. They also lock up the BOINC software, other tasks do not run or download. When I abort all the 6 cpu tasks the other projects immediately download and run normally.

I've just aborted a work unit. 6+ hours CPU, only 8% complete, deadline later today. Your work units normally execute in a few minutes on here, (4GHz i7 + GPU). All other units aborted also before they start, NNT set. Something is wrong now.

Eric,
My multi-thread work unit runs slow like most of the cases listed here. I noticed on the windows "task manager" (press Ctrl-Alt-Del), on the performance tab, that the work units are only using less then 20% of the CPU time. It is not a setting in my app. All of the other projects use 80-100% of the CPU time.
Michael

I am running W10,. If the multi thread tasks are the ones using 6 cpus I am having trouble. The tasks run very slowly, 3% after two days.

I am running 6 cores on Win7 64-bit. On an i7-4771 they are taking about 4 to 6 hours now (one is estimating 8 hours), but completing OK at this point.

Is there a fix for the 6CPU task problem? I just aborted 19 tasks so the other projects could run. I may have to drop MW and look for another project. Is there a way to download just the single CPU tasks?

Is there a fix for the 6CPU task problem?

They work for me.
https://milkyway.cs.rpi.edu/milkyway/results.php?hostid=737912&offset=0&show_names=0&state=4&appid=
Are you sure it is not your AV interfering?